Intelligent power grid key node situation awareness method based on multi-dimensional index dynamic fusion operator
By building a multi-parameter collaborative analysis model and node state evaluation model, combining pattern recognition algorithms and network topology correlation evaluation, the problem that situational awareness methods of key nodes of smart grids are difficult to integrate multi-dimensional indicators and adapt to grid changes is solved, and high-precision situational awareness and prediction are achieved, which improves the safety and reliability of grid operation.
Patent Information
- Application Number
- CN202411610877.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-12
- Publication Date
- 2025-03-14
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The situational awareness methods of existing smart grid key nodes are difficult to effectively integrate multi-dimensional indicator information, dynamically adapt to changes in the power grid operating environment, and it is difficult to achieve accurate status recognition and prediction under limited historical data conditions.
Using a method based on the dynamic fusion operator of multi-dimensional indicators, a multi-parameter collaborative analysis model and node state evaluation model are constructed, and precise perception and prediction of the situation of key nodes of smart grids are achieved through pattern recognition algorithms and network topology correlation evaluation.
It improves the perception accuracy and early warning timeliness of key node states of smart grids, enhances the ability to adapt to dynamic changes of the power grid, provides a more reliable decision-making basis, reduces the risk of failure, and optimizes system performance.
Smart Images
Figure CN119622401A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of power system monitoring and management technology, mainly to the fields of real-time monitoring, fault warning, operation optimization, etc. of smart grids, providing important decision support for the safe and stable operation of power grids, and specifically to a smart grid key node situation awareness method based on a multi-dimensional indicator dynamic fusion operator. Background Art
[0002] In modern power systems, smart grids, as an important national infrastructure, play a vital role in ensuring economic development, social stability and people's quality of life. The operating status of smart grids directly affects industrial production, urban operations, residential electricity consumption and many other aspects. Its safety and reliability are directly related to national energy security and the sustainable development of the economy and society. With the continuous expansion of the scale of power grids and the increasing complexity of their structures, accurately grasping the operating status of key nodes in smart grids and timely predicting and preventing possible failures are of great significance for improving the operating efficiency of power grids, reducing economic losses and ensuring power supply safety. Therefore, developing advanced smart grid key node situation awareness technology to achieve real-time monitoring of power grid operation status, accurate early warning and scientific decision-making has become an important topic in the current power system research and application.
[0003] At present, there are two main methods for situational awareness of key nodes in smart grids: the first is the traditional threshold-based monitoring method. This method determines the node status by setting thresholds for various indicators, such as voltage, current, power factor, etc. When an indicator exceeds the preset threshold, the system will issue an alarm. This method is simple to implement and has a small amount of calculation, but it has obvious defects. First, due to the complex and changeable operating environment of the power grid, fixed thresholds are difficult to adapt to all situations, and false alarms or missed alarms are prone to occur; secondly, this method is difficult to capture the mutual influence and potential correlation between multiple indicators, and may ignore some complex abnormal states. The second type is a prediction method based on artificial intelligence, such as using algorithms such as neural networks or support vector machines to predict node status. This type of method can handle more complex data relationships and has high prediction accuracy. However, they also face some challenges: first, these methods usually require a large amount of labeled data for training, while in actual power grid operation, abnormal state data is relatively scarce, and it is difficult to obtain sufficient training samples; second, due to the dynamic changes in power grid structure and load characteristics, the generalization ability of the model is challenged and may perform poorly when faced with new and unseen states; finally, the decision-making process of many artificial intelligence models lacks explainability, which may be limited in power systems that require high transparency and reliability.
[0004] Therefore, we urgently need a new smart grid key node situation awareness method that can effectively integrate multi-dimensional indicator information, dynamically adapt to changes in the grid operation environment, and achieve accurate state identification and prediction under limited historical data conditions. This method should be able to overcome the limitations of existing technologies, improve the perception accuracy and early warning timeliness of the key node status of the grid, and thus provide strong support for the safe operation and optimized management of the grid. Summary of the invention
[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method for situation awareness of key nodes of smart grids based on a multi-dimensional indicator dynamic fusion operator. The method constructs a multi-parameter collaborative analysis model, which can adaptively integrate multiple power system characteristic quantities such as voltage, current, and power, and identify the change patterns of these characteristic quantities on multiple time scales; by introducing innovative pattern recognition algorithms, the method can effectively identify and classify different node operating states without the need for manual labeling of data. Furthermore, a node state evaluation model is constructed using a correlation evaluation method based on network topology; finally, through the combination of a multi-parameter collaborative analysis model and a state evaluation model, accurate perception and prediction of the situation of key nodes of smart grids can be achieved, thereby improving the safety and reliability of grid operation.
[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:
[0007] A method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator includes the following steps:
[0008] S1, input the power situation database PowerStateBase based on the multidimensional indicator data set of key nodes of the smart grid, and obtain the number of features of the data set, the number of samples of the data set, the number of data layers, the key information density threshold, and the random sampling ratio, and establish the power grid situation pyramid GridStatePyramid;
[0009] S2, establishing a power feature extractor PowerFeatureExtractor, inputting the power grid state pyramid GridStatePyramid, and obtaining a power feature set PowerFeatureSet;
[0010] S3, establish a power information concentrator PowerInfoConcentrator, input the power feature set PowerFeatureSet obtained in S2, and obtain the optimized power feature OptimizedPowerFeature;
[0011] S4, combining the power feature extractor PowerFeatureExtractor and the power information concentrator PowerInfoConcentrator, constructs and trains the grid situation sensor GridStatePerceptor for situation awareness of key nodes in the smart grid;
[0012] S5, using the trained grid situation perception GridStatePerceptor to conduct situation perception of the new smart grid key node data.
[0013] Furthermore, the number of features of the data set in step S1 is the power grid index space ElectricIndexSpace, the number of samples of the data set is the power situation points PowerStateCount, the number of data layers is the power grid layer number GridLayerNum, the key information density threshold is the power information density threshold PowerInfoThreshold, and the random sampling ratio is the power grid sampling rate GridSampleRate.
[0014] Furthermore, the specific steps of establishing the grid state pyramid GridStatePyramid in step S1 are as follows:
[0015] S101, inputting a power situation database PowerStateBase including multi-dimensional power indicators;
[0016] S102, obtaining the dimension of the power grid index space ElectricIndexSpace according to the dimension of the power situation database PowerStateBase; obtaining the power situation point PowerStateCount according to the number of samples of the power situation database PowerStateBase;
[0017] S103, establishing a power grid state pyramid GridStatePyramid having a GridLayerNum level data structure, and initializing each level thereof to be empty;
[0018] S104, initializing the pyramid level index GridLayerIndex=1;
[0019] S105, let the current grid layer data CurrentGridLayerData = power state database PowerStateBase;
[0020] S106, set the current layer state points CurrentLayerStateCount=power state points PowerStateCount;
[0021] S107, calculating the current layer power density CurrentLayerPowerDensity;
[0022] S108, if the current layer power density CurrentLayerPowerDensity < the power information density threshold PowerInfoThreshold or the grid layer index GridLayerIndex > the number of layers GridLayerNum, go to S112, otherwise go to step S109;
[0023] S109, executing a probability screening operation, performing random sampling on the current grid layer data CurrentGridLayerData with a probability of the grid sampling rate GridSampleRate, and obtaining grid filtered data GridFilteredData;
[0024] S110, let the data in the layer corresponding to the current layer index in the power grid situation pyramid GridStatePyramid[GridLayerIndex] = power grid filtered data GridFilteredData;
[0025] S111, set the pyramid level index GridLayerIndex = GridLayerIndex + 1; update the current grid layer data CurrentGridLayerData = grid filter data GridFilteredData; go to S106;
[0026] S112, let GridStatePyramid[GridLayerIndex]=current grid layer data CurrentGridLayerData;
[0027] S113, end the operation.
[0028] Furthermore, in step S2, a power feature extractor PowerFeatureExtractor is established, and the power grid state pyramid GridStatePyramid is input to obtain the power feature set PowerFeatureSet. The specific steps are as follows:
[0029] S201, initialize the power feature set PowerFeatureSet = empty vector;
[0030] S202, set the grid layer traverser GridLayerIterator = 1;
[0031] S203, let the current layer power data CurrentLayerPowerData = the data corresponding to the grid layer traverser in the grid situation pyramid GridStatePyramid[GridLayerIterator];
[0032] S204, calculating the current layer power feature CurrentLayerPowerFeature;
[0033] S205, connecting the obtained current layer power feature CurrentLayerPowerFeature and the power feature set PowerFeatureSet to obtain the power feature set PowerFeatureSet;
[0034] S206, set the grid layer traverser GridLayerIterator = GridLayerIterator + 1;
[0035] S207, if GridLayerIterator<=GridLayerNum, go to S203; otherwise go to S208;
[0036] S208, outputting the power feature set PowerFeatureSet as the result extracted by the power feature extractor PowerFeatureExtractor.
[0037] Furthermore, the calculation of the current layer power feature CurrentLayerPowerFeature in step S204 includes: calculating basic statistical features, calculating a voltage stability index, calculating a load balancing factor, calculating a power factor, calculating a harmonic distortion factor, and calculating a frequency stability index.
[0038] Furthermore, in step S3, the specific steps of establishing a power information concentrator PowerInfoConcentrator, inputting a power feature set PowerFeatureSet, and obtaining an optimized power feature OptimizedPowerFeature are as follows:
[0039] S301, initializing optimized power feature OptimizedPowerFeature = power feature set PowerFeatureSet;
[0040] S302, calculating the initial power density InitialPowerDensity;
[0041] S303, setting the power optimization iteration number PowerOptimizeIter=1;
[0042] S304, performing feature selection operation to obtain a temporary power feature TempPowerFeature;
[0043] S305, calculating a temporary power density TempPowerDensity;
[0044] S306, if the temporary power density TempPowerDensity>the initial power density InitialPowerDensity, then set the optimized power feature OptimizedPowerFeature=temporary power feature TempPowerFeature and the initial power density InitialPowerDensity=temporary power density TempPowerDensity;
[0045] S307, setting the power optimization iteration number PowerOptimizeIter=PowerOptimizeIter+1;
[0046] S308, if the power optimization iteration number PowerOptimizeIter is less than the preset maximum optimization number and |TempPowerDensity-InitialPowerDensity|>1e-6, go to S304; otherwise go to S309;
[0047] S309, outputting the optimized power feature OptimizedPowerFeature as a result of the power information concentrator PowerInfoConcentrator.
[0048] Furthermore, step S304 performs a feature selection operation, and the specific steps of obtaining the temporary power feature TempPowerFeature are as follows:
[0049] S3041, calculating the power correlation coefficient and power mutual information of each power feature;
[0050] S3042, let power feature weight PowerFeatureWeight = |power correlation coefficient|*power mutual information;
[0051] S3041, normalizing the power feature weight PowerFeatureWeight;
[0052] S3044, remove the 10% features with the lowest power feature weight PowerFeatureWeight;
[0053] S3045, perform power principal component analysis on the remaining features;
[0054] S3046, select the principal component that explains 90% of the variance as the temporary power feature TempPowerFeature.
[0055] Furthermore, the specific steps of constructing the grid state sensor GridStatePerceptor for smart grid key node state awareness in step S4 are as follows:
[0056] S401, input data: input power training data PowerTrainingData;
[0057] S402, extracting features: using a power feature extractor PowerFeatureExtractor to extract features from the power training data PowerTrainingData to obtain a power training feature PowerTrainingFeature;
[0058] S403, optimizing features: using a power information concentrator PowerInfoConcentrator to process the power training feature PowerTrainingFeatur to obtain an optimized power feature OptimizedPowerFeature;
[0059] S404, model establishment and training: construct a situation awareness model GridStatePerceptor, and input the optimized power feature OptimizedPowerFeature for training using the cross-validation method to obtain a trained power grid situation awareness model GridStatePerceptor.
[0060] Furthermore, the specific steps of using the trained grid situation sensor GridStatePerceptor to perform situation awareness on the new smart grid key node data described in S5 are as follows:
[0061] S501, inputting new smart grid key node data to be analyzed, namely, new state data NewGridStateData;
[0062] S502, using a power feature extractor PowerFeatureExtractor to extract features from the new state data NewGridStateData to obtain a new power feature NewPowerFeature;
[0063] S503, using a power information concentrator PowerInfoConcentrator to process the new power feature NewPowerFeature to obtain an optimized new power feature OptimizedNewPowerFeatur;
[0064] S504, inputting the optimized new power feature OptimizedNewPowerFeatur into the trained grid situation perception GridStatePerceptor to obtain the situation perception result GridStateOutput output by the model.
[0065] Furthermore, the situation awareness result GridStateOutput includes the probability distribution of the grid state, key influencing factors and their importance, and gives corresponding processing suggestions according to different grid states.
[0066] The remarkable effects of the present invention are:
[0067] 1. The method described in the present invention constructs a multi-parameter collaborative analysis model, which can adaptively integrate multiple power system characteristic quantities such as voltage, current, power, and identify the change patterns of these characteristic quantities on multiple time scales; then by introducing an innovative pattern recognition algorithm, the method described in the present invention can effectively identify and classify different node operating states without manually labeling data; then, a node state evaluation model is constructed using a correlation evaluation method based on network topology; finally, through the combination of the multi-parameter collaborative analysis model and the state evaluation model, accurate perception and prediction of the status of key nodes in the smart grid can be achieved, thereby improving the safety and reliability of grid operation.
[0068] 2. The use of the patent of this invention can break through the limitations of the traditional fixed threshold method, and accurately grasp the operating status of key nodes of the smart grid by analyzing the spatiotemporal variation characteristics of the characteristic quantity of the power system, which not only improves the accuracy and real-time performance of situation awareness, but also enhances the adaptability to dynamic changes in the power grid. Therefore, the present invention provides a more reliable decision-making basis for power grid dispatching and management, helps to reduce the risk of failure, optimize system performance, and thus make important contributions to ensuring the stability of power supply and promoting energy efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0069] Figure 1 is a flow chart of the method of the present invention;
[0070] Figure 2 This is the performance evaluation result diagram of the grid fault predictor GridFaultPredictor model; DETAILED DESCRIPTION
[0071] The specific implementation manner and working principle of the present invention are further described in detail below with reference to the accompanying drawings.
[0072] This embodiment demonstrates the application of a smart grid key node situation awareness method based on multi-dimensional index dynamic fusion operator in fault prediction. We take a 330kV substation of a provincial power company as an example to demonstrate the specific implementation steps and effects of this technology.
[0073] like Figure 1 As shown in FIG, a method for situation awareness of key nodes of a smart grid based on a dynamic fusion operator of multi-dimensional indicators is described, and the specific steps are as follows:
[0074] S1, input the power situation database PowerStateBase based on the multidimensional indicator data set of key nodes of the smart grid, and obtain the number of features of the data set, the number of samples of the data set, the number of data layers, the key information density threshold, and the random sampling ratio, and establish the power grid situation pyramid GridStatePyramid;
[0075] In this step, the number of features of the data set is the power grid index space ElectricIndexSpace, the number of samples of the data set is the power situation points PowerStateCount, the number of data layers is the power grid layer number GridLayerNum, the key information density threshold is the power information density threshold PowerInfoThreshold, and the random sampling ratio is the power grid sampling rate GridSampleRate.
[0076] Therefore, the specific steps to establish the grid state pyramid GridStatePyramid are as follows:
[0077] S101, input the power situation database PowerStateBase, which contains the operation data of the substation within one month, a total of 43,200 records (one record per minute), each record includes 28 power indicators such as voltage, current, active power, reactive power, frequency, etc., as shown in Table 1:
[0078] Table 1: Example table of some data in the power situation database PowerStateBase
[0079]
[0080] S102, set the dimension of the power situation database PowerStateBase = the dimension of the power grid index space ElectricIndexSpace, that is, the dimension of the power grid index space ElectricIndexSpace = 28; obtain the power situation points PowerStateCount according to the number of samples of the power situation database PowerStateBase, that is, the power situation points PowerStateCount = 43,200;
[0081] S103, establishing a grid state pyramid GridStatePyramid having a GridLayerNum=5-level data structure, and initializing each level thereof to be empty;
[0082] S104, initializing the pyramid level index GridLayerIndex=1;
[0083] S105, let the current grid layer data CurrentGridLayerData = power state database PowerStateBase;
[0084] S106, set the current layer state points CurrentLayerStateCount=power state points PowerStateCount=43,200;
[0085] S107, calculate the current layer power density CurrentLayerPowerDensity, the calculation formula is:
[0086] CurrentLayerPowerDensity=(log(1+standard deviation(CurrentGridLayerData))*power entropy(CurrentGridLayerData)) / (CurrentLayerStateCount*log(ElectricIndexSpace)),
[0087] Wherein, power entropy (X) = -∑(p(x)*log(p(x))), p(x) is the probability of each unique value in X;
[0088] S108, if the current layer power density CurrentLayerPowerDensity<0.01 or the grid layer index GridLayerIndex>5, go to S112, otherwise go to step S109;
[0089] S109, executing a probability screening operation, performing random sampling on the current grid layer data CurrentGridLayerData with a probability of the grid sampling rate GridSampleRate being 0.5, and obtaining the grid filtered data GridFilteredData;
[0090] S110, let the data in the layer corresponding to the current layer index in the power grid situation pyramid GridStatePyramid[GridLayerIndex] = power grid filtered data GridFilteredData;
[0091] S111, set the pyramid level index GridLayerIndex = GridLayerIndex + 1; update the current grid layer data CurrentGridLayerData = grid filter data GridFilteredData; go to S106;
[0092] S112, let GridStatePyramid[GridLayerIndex]=current grid layer data CurrentGridLayerData;
[0093] S113, end the operation.
[0094] S2, establish a power feature extractor PowerFeatureExtractor, input the power grid state pyramid GridStatePyramid, and obtain the power feature set PowerFeatureSet. The specific steps are as follows:
[0095] S201, initialize the power feature set PowerFeatureSet = empty vector;
[0096] S202, set the grid layer traverser GridLayerIterator = 1;
[0097] S203, let the current layer power data CurrentLayerPowerData = the data corresponding to the grid layer traverser in the grid situation pyramid GridStatePyramid[GridLayerIterator];
[0098] S204, calculate the current layer power feature CurrentLayerPowerFeature, specifically:
[0099] S2041, calculate basic statistical characteristics: mean, variance, skewness, and kurtosis of voltage, current, and power;
[0100] S2042, calculating voltage stability index VSI=1-(maximum voltage deviation / rated voltage);
[0101] S2043, calculate the load balancing factor LBF = 1-(maximum load-minimum load) / average load;
[0102] S2044, calculate the power factor PF = active power / √(active power^2+reactive power^2);
[0103] S2045, calculate the harmonic distortion factor THD = √(Σharmonic component^2) / fundamental wave component;
[0104] S2046, calculate frequency stability index FSI = 1-|actual frequency-rated frequency| / rated frequency;
[0105] S205, connecting the obtained current layer power feature CurrentLayerPowerFeature and the power feature set PowerFeatureSet to obtain the power feature set PowerFeatureSet;
[0106] S206, set the grid layer traverser GridLayerIterator = GridLayerIterator + 1;
[0107] S207, if GridLayerIterator<=5, go to S203; otherwise go to S208;
[0108] S208, outputting the power feature set PowerFeatureSet as the result extracted by the power feature extractor PowerFeatureExtractor.
[0109] S3, establish a power information concentrator PowerInfoConcentrator, input the power feature set PowerFeatureSet obtained in S2, and obtain the optimized power feature OptimizedPowerFeature. The specific steps are as follows:
[0110] S301, initializing optimized power feature OptimizedPowerFeature = power feature set PowerFeatureSet;
[0111] S302, calculate the initial power density InitialPowerDensity, the calculation formula is:
[0112] InitialPowerDensity=(H(OptimizedPowerFeature)*I(OptimizedPowerFeature;
[0113] PowerFeatureSet)) / log(size(OptimizedPowerFeature)),
[0114] Among them, H(X) is the power entropy of X, I(X;Y) is the power mutual information of X and Y;
[0115] S303, setting the power optimization iteration number PowerOptimizeIter=1;
[0116] S304, performing feature selection operation to obtain temporary power feature TempPowerFeature, as follows:
[0117] S3041, calculating the power correlation coefficient and power mutual information of each power feature;
[0118] S3042, let power feature weight PowerFeatureWeight = |power correlation coefficient|*power mutual information;
[0119] S3041, normalizing the power feature weight PowerFeatureWeight;
[0120] S3044, remove the 10% features with the lowest power feature weight PowerFeatureWeight;
[0121] S3045, perform power principal component analysis on the remaining features;
[0122] S3046, select the principal component that explains 90% of the variance as the temporary power feature TempPowerFeature;
[0123] S305, calculate the temporary power density TempPowerDensity, the calculation formula is:
[0124] TempPowerDensity=H(TempPowerFeature)*I(TempPowerFeature;PowerFeatureSet)) / log(size(TempPowerFeature));
[0125] S306, if the temporary power density TempPowerDensity>the initial power density InitialPowerDensity, then set the optimized power feature OptimizedPowerFeature=temporary power feature TempPowerFeature and the initial power density InitialPowerDensity=temporary power density TempPowerDensity;
[0126] S307, setting the power optimization iteration number PowerOptimizeIter=PowerOptimizeIter+1;
[0127] S308, if the power optimization iteration number PowerOptimizeIter is less than the preset maximum optimization number of 100 and |TempPowerDensity-InitialPowerDensity|>1e-6, go to S304; otherwise go to S309;
[0128] S309, outputting the optimized power feature OptimizedPowerFeature as a result of the power information concentrator PowerInfoConcentrator.
[0129] The optimized power feature OptimizedPowerFeature obtained in this embodiment is shown in Table 2:
[0130] Table 2: The first 10 features in the optimized power feature OptimizedPowerFeature
[0131]
[0132] S4, combining the power feature extractor PowerFeatureExtractor and the power information concentrator PowerInfoConcentrator, constructs and trains the grid situation sensor GridStatePerceptor for situation awareness of key nodes of the smart grid. The specific steps are as follows:
[0133] S401, input data: input power training data PowerTrainingData, wherein the power training data PowerTrainingData is 80% of the data in the power situation database PowerStateBase (a total of 34,560 records);
[0134] S402, extracting features: using a power feature extractor PowerFeatureExtractor to extract features from the power training data PowerTrainingData to obtain a power training feature PowerTrainingFeature;
[0135] S403, optimizing features: using a power information concentrator PowerInfoConcentrator to process the power training feature PowerTrainingFeatur to obtain an optimized power feature OptimizedPowerFeature;
[0136] S404, model building and training: construct the situation awareness model GridStatePerceptor, and input the optimized power feature OptimizedPowerFeature to train using the cross-validation method to obtain the trained power grid situation awareness model GridStatePerceptor:
[0137] S4041, using the optimized power feature OptimizedPowerFeature as an input feature;
[0138] S4042, uses an ensemble learning approach combining decision trees, random forests, and gradient boosted trees;
[0139] S4043, optimize model parameters using 5-fold cross validation;
[0140] S4044, the trained model is the grid fault prediction model GridFaultPredictor required in this embodiment.
[0141] The performance evaluation results of the power grid fault prediction model GridFaultPredictor obtained in this example are as follows Figure 2 shown.
[0142] S5, using the trained grid situation perception model, that is, the grid fault prediction model GridStatePerceptor, to perform situation perception on the new smart grid key node data. The specific steps are as follows:
[0143] S501, inputting new smart grid key node data to be analyzed, namely new state data NewGridStateData. In this example, the new state data NewGridStateData uses the remaining 20% of the data (a total of 8,640 records) in the power state database PowerStateBase;
[0144] S502, using a power feature extractor PowerFeatureExtractor to extract features from the new state data NewGridStateData to obtain a new power feature NewPowerFeature;
[0145] S503, using a power information concentrator PowerInfoConcentrator to process the new power feature NewPowerFeature to obtain an optimized new power feature OptimizedNewPowerFeatur;
[0146] S504, inputting the optimized new power feature OptimizedNewPowerFeatur into the trained grid situation perception GridStatePerceptor to obtain the situation perception result GridStateOutput output by the model, including:
[0147] 1) The probability of failure occurring in the next 24 hours;
[0148] 2) Output the key factors that may cause failure and their importance;
[0149] 3) Provide corresponding processing suggestions based on the predicted failure probability, such as:
[0150] - Low risk of failure (<10%): Continue routine monitoring;
[0151] - Medium failure risk (10%-50%): Increase monitoring frequency and prepare for preventive maintenance;
[0152] - High risk of failure (>50%): immediately conduct a comprehensive inspection and take preventive measures.
[0153] The fault prediction results obtained in step S5 of this embodiment are shown in Table 3:
[0154] Table 3: Example of fault prediction results
[0155]
[0156] By applying the smart grid key node situation awareness method of the present invention to predict faults, we achieved the following results on the test data set:
[0157] 1. Fault prediction accuracy: 89%;
[0158] 2. Fault warning advance time: 6 hours on average;
[0159] 3. False positive rate: 8%;
[0160] 4. False negative rate: 3%;
[0161] Case Study:
[0162] During the period of 02:00-03:00 on May 25, the system predicted a high failure risk (failure probability 75%), mainly due to abnormal harmonic content. Based on this prediction, the operation and maintenance personnel immediately conducted an inspection and found a transformer that was about to fail. They repaired it in time and avoided a possible large-scale power outage.
[0163] Therefore, the smart grid key node situation awareness method described in the present invention shows excellent performance in fault prediction. It can effectively integrate multi-dimensional indicator information and accurately predict potential faults of substations without the need for a large amount of manual annotation data. This method not only improves the accuracy and timeliness of fault prediction, but also provides actionable suggestions for operation and maintenance personnel, effectively reduces the risk of equipment failure, and improves the safety and reliability of the power grid.
[0164] In summary, the method described in the present invention constructs a multi-parameter collaborative analysis model, which can adaptively integrate multiple power system characteristic quantities such as voltage, current, power, and identify the change patterns of these characteristic quantities on multiple time scales; then, by introducing an innovative pattern recognition algorithm, the method described in the present invention can effectively identify and classify different node operating states without manually labeling data; then, a node state evaluation model is constructed by using a correlation evaluation method based on network topology; finally, through the combination of the multi-parameter collaborative analysis model and the state evaluation model, accurate perception and prediction of the status of key nodes of the smart grid can be achieved, thereby improving the safety and reliability of grid operation.
[0165] The technical solution provided by the present invention is described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, the present invention can also be improved and modified in a number of ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.
Claims
1. A method for situation awareness of key nodes of smart grid based on multi-dimensional indicator dynamic fusion operator, characterized in that: The steps include: S1, input the power situation database PowerStateBase based on the multi-dimensional indicator data set of key nodes of the smart grid, and obtain the number of features of the data set, the number of samples of the data set, the number of data layers, the key information density threshold, and the random sampling ratio, and establish the power grid situation pyramid GridStatePyramid; S2, establishing a power feature extractor PowerFeatureExtractor, inputting the power grid state pyramid GridStatePyramid, and obtaining a power feature set PowerFeatureSet; S3, establish a power information concentrator PowerInfoConcentrator, input the power feature set PowerFeatureSet obtained in S2, and obtain the optimized power feature OptimizedPowerFeature; S4, combining the power feature extractor PowerFeatureExtractor and the power information concentrator PowerInfoConcentrator, constructs and trains the grid situation sensor GridStatePerceptor for situation awareness of key nodes in the smart grid; S5, using the trained grid situation perception GridStatePerceptor to conduct situation perception of the new smart grid key node data.
2. The method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator according to claim 1 is characterized in that: The number of features of the data set in step S1 is the power grid index space ElectricIndexSpace, the number of samples of the data set is the power situation points PowerStateCount, the number of data layers is the power grid layer number GridLayerNum, the key information density threshold is the power information density threshold PowerInfoThreshold, and the random sampling ratio is the power grid sampling rate GridSampleRate.
3. The method for situation awareness of key nodes of smart grid based on multi-dimensional indicator dynamic fusion operator according to claim 2 is characterized in that: The specific steps of establishing the grid state pyramid GridStatePyramid in step S1 are as follows: S101, inputting a power situation database PowerStateBase including multi-dimensional power indicators; S102, obtaining the dimension of the power grid index space ElectricIndexSpace according to the dimension of the power situation database PowerStateBase; obtaining the power situation point PowerStateCount according to the number of samples of the power situation database PowerStateBase; S103, establishing a power grid state pyramid GridStatePyramid having a GridLayerNum level data structure, and initializing each level thereof to be empty; S104, initializing the pyramid level index GridLayerIndex=1; S105, let the current grid layer data CurrentGridLayerData = power state database PowerStateBase; S106, set the current layer state points CurrentLayerStateCount=power state points PowerStateCount; S107, calculating the current layer power density CurrentLayerPowerDensity; S108, if the current layer power density CurrentLayerPowerDensity < the power information density threshold PowerInfoThreshold or the grid layer index GridLayerIndex > the number of layers GridLayerNum, go to S112, otherwise go to step S109; S109, executing a probability screening operation, performing random sampling on the current grid layer data CurrentGridLayerData with a probability of the grid sampling rate GridSampleRate, and obtaining grid filtered data GridFilteredData; S110, let the data in the layer corresponding to the current layer index in the power grid situation pyramid GridStatePyramid[GridLayerIndex] = power grid filtered data GridFilteredData; S111, set the pyramid level index GridLayerIndex = GridLayerIndex + 1; update the current grid layer data CurrentGridLayerData = grid filter data GridFilteredData; go to S106; S112, let GridStatePyramid[GridLayerIndex]=current grid layer data CurrentGridLayerData; S113, end the operation.
4. The method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator according to claim 1 is characterized in that: In step S2, the specific steps of establishing a power feature extractor PowerFeatureExtractor, inputting the power grid state pyramid GridStatePyramid, and obtaining the power feature set PowerFeatureSet are as follows: S201, initialize the power feature set PowerFeatureSet = empty vector; S202, set the grid layer traverser GridLayerIterator = 1; S203, let the current layer power data CurrentLayerPowerData = the data corresponding to the grid layer traverser in the grid situation pyramid GridStatePyramid[GridLayerIterator]; S204, calculating the current layer power feature CurrentLayerPowerFeature; S205, connecting the obtained current layer power feature CurrentLayerPowerFeature and the power feature set PowerFeatureSet to obtain the power feature set PowerFeatureSet; S206, set the grid layer traverser GridLayerIterator = GridLayerIterator + 1; S207, if GridLayerIterator<=GridLayerNum, go to S203; otherwise go to S208; S208, outputting the power feature set PowerFeatureSet as the result extracted by the power feature extractor PowerFeatureExtractor.
5. The method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator according to claim 4 is characterized in that: The calculation of the current layer power feature CurrentLayerPowerFeature in step S204 includes: calculating basic statistical features, calculating voltage stability index, calculating load balancing factor, calculating power factor, calculating harmonic distortion factor, and calculating frequency stability index.
6. The method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator according to claim 1 is characterized in that: In step S3, the specific steps of establishing a power information concentrator PowerInfoConcentrator, inputting a power feature set PowerFeatureSet, and obtaining an optimized power feature OptimizedPowerFeature are as follows: S301, initializing optimized power feature OptimizedPowerFeature = power feature set PowerFeatureSet; S302, calculating the initial power density InitialPowerDensity; S303, setting the power optimization iteration number PowerOptimizeIter=1; S304, performing feature selection operation to obtain a temporary power feature TempPowerFeature; S305, calculating a temporary power density TempPowerDensity; S306, if the temporary power density TempPowerDensity>the initial power density InitialPowerDensity, then set the optimized power feature OptimizedPowerFeature=temporary power feature TempPowerFeature and the initial power density InitialPowerDensity=temporary power density TempPowerDensity; S307, setting the power optimization iteration number PowerOptimizeIter=PowerOptimizeIter+1; S308, if the power optimization iteration number PowerOptimizeIter is less than the preset maximum optimization number and |TempPowerDensity-InitialPowerDensity|>1e-6, go to S304; otherwise go to S309; S309, outputting the optimized power feature OptimizedPowerFeature as a result of the power information concentrator PowerInfoConcentrator.
7. The method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator according to claim 6 is characterized in that: Step S304 performs feature selection operation, and the specific steps of obtaining the temporary power feature TempPowerFeature are as follows: S3041, calculating the power correlation coefficient and power mutual information of each power feature; S3042, let power feature weight PowerFeatureWeight = |power correlation coefficient|*power mutual information; S3041, normalizing the power feature weight PowerFeatureWeight; S3044, remove the 10% features with the lowest power feature weight PowerFeatureWeight; S3045, perform power principal component analysis on the remaining features; S3046, select the principal component that explains 90% of the variance as the temporary power feature TempPowerFeature.
8. The method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator according to claim 1 is characterized in that: The specific steps of constructing the grid state sensor GridStatePerceptor for smart grid key node situation awareness described in step S4 are as follows: S401, input data: input power training data PowerTrainingData; S402, extracting features: using a power feature extractor PowerFeatureExtractor to extract features from the power training data PowerTrainingData to obtain a power training feature PowerTrainingFeature; S403, optimizing features: using a power information concentrator PowerInfoConcentrator to process the power training feature PowerTrainingFeatur to obtain an optimized power feature OptimizedPowerFeature; S404, model establishment and training: construct a situation awareness model GridStatePerceptor, and input the optimized power feature OptimizedPowerFeature to perform training using the cross-validation method to obtain a trained power grid situation awareness model GridStatePerceptor.
9. The method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator according to claim 1 is characterized in that: The specific steps of using the trained grid state perceiver GridStatePerceptor to perform situation awareness on the new smart grid key node data described in S5 are as follows: S501, inputting new smart grid key node data to be analyzed, namely, new state data NewGridStateData; S502, using a power feature extractor PowerFeatureExtractor to extract features from the new state data NewGridStateData to obtain a new power feature NewPowerFeature; S503, using a power information concentrator PowerInfoConcentrator to process the new power feature NewPowerFeature to obtain an optimized new power feature OptimizedNewPowerFeatur; S504, inputting the optimized new power feature OptimizedNewPowerFeatur into the trained grid situation perception GridStatePerceptor to obtain the situation perception result GridStateOutput output by the model.
10. The method for situation awareness of key nodes of a smart grid based on a multi-dimensional indicator dynamic fusion operator according to claim 9 is characterized in that: The situation awareness result GridStateOutput includes the probability distribution of the grid state, key influencing factors and their importance, and corresponding processing suggestions are given according to different grid states.